Exploiting Hidden-Layer Responses of Deep Neural Networks for Language Recognition

• https://www.researchgate.net/publication/307889648_Exploiting_Hidden-Layer_Responses_of_Deep_Neural_Networks_for_Language_Recognition
• http://www.fit.vutbr.cz/research/groups/speech/publi/2016/li_interspeech2016_IS161584.pdf PDF

LID, I-vector, DNN, hidden layers

The most popular way to apply Deep Neural Network (DNN) for Language IDentification (LID) involves the extraction of bottleneck features from a network that was trained on automatic speech recognition task. These features are modeled using a classical I-vector system. Recently, a more direct DNN approach was proposed, it consists of estimating the language posteriors directly from a stacked frames input. The final decision score is based on averaging the scores for all the frames for a given speech segment. In this paper, we extended the direct DNN approach by modeling all hidden-layer activations rather than just averaging the output scores. One super-vector per utterance is formed by concatenating all hidden-layer responses. The dimensionality of this vector is then reduced using a Principal Component Analysis (PCA). The obtained reduce vector summarizes the most discriminative features for language recognition based on the trained DNNs. We evaluated this approach in NIST 2015 language recognition evaluation. The performances achieved by the proposed approach are very competitive to the classical I-vector baseline.

The most popular way to apply Deep Neural Network (DNN) for Language IDentification (LID) involves the extraction of bottleneck features from a network that was trained on automatic speech recognition task. These features are modeled using a classical I-vector system. Recently, a more direct DNN approach was proposed, it consists of estimating the language posteriors directly from a stacked frames input. The final decision score is based on averaging the scores for all the frames for a given speech segment. In this paper, we extended the direct DNN approach by modeling all hidden-layer activations rather than just averaging the output scores. One super-vector per utterance is formed by concatenating all hidden-layer responses. The dimensionality of this vector is then reduced using a Principal Component Analysis (PCA). The obtained reduce vector summarizes the most discriminative features for language recognition based on the trained DNNs. We evaluated this approach in NIST 2015 language recognition evaluation. The performances achieved by the proposed approach are very competitive to the classical I-vector baseline.